AI and Networksedit

AI and Networks is a primary research direction of Xinbao Qiao. It covers AI for Networks, Networks for AI, and learning algorithms whose behavior depends on communication, decentralization, edge devices, institutional data silos, or cross-party evaluation.¹

Introductionedit

AI and networks research examines how learning is shaped by where data live, how information moves, and which parties can evaluate a model. The topic therefore includes AI-assisted networked systems, network support for AI systems, decentralized learning, distributed computing, data pruning, collaborative evaluation, and synthetic-data verification under low-resource or siloed access.

Research contextedit

Qiao's background in communication engineering connects with his later work on distributed learning, data silos, collaborative evaluation, distributed Wasserstein barycenters, and data pruning for decentralized training. These problems treat communication, locality, and infrastructure as factors that shape learning.

Current doctoral focusedit

In the CUHK doctoral stage, Qiao's recent work within this topic centers on data-centric ML, AI for Networks, and Networks for AI. The emphasis is on learning systems whose data and evaluation evidence are shaped by communication, network infrastructure, and decentralized access, rather than on centralized statistical objectives alone.

Publicationsedit

Paper	Venue/status
When Sample Selection Bias Precipitates Model Collapse	ICML 2026, 6-11 July 2026, Seoul.

Manuscripts under reviewedit

The under-review manuscript Decentralized Free-Support Wasserstein Barycenter develops this direction through free-support barycenter computation: learning support locations lets the shared reference adapt to distributional geometry, while separating the communication needed for barycenter quality from that needed for network agreement.

Connection to Qiao's workedit

The ICML 2026 paper When Sample Selection Bias Precipitates Model Collapse studies collaborative verification when low-resource parties only hold fragmented local evidence. Distributed Wasserstein barycenter methods remain one technical route for treating a reference distribution as something computed across a network rather than assumed to exist centrally. Earlier work in machine unlearning reflects the same systems perspective: algorithms are evaluated not only by accuracy, but also by latency, communication, and the cost of changing data after training.

See alsoedit

• Distributed Learning
• Data Silos
• Collaborative Evaluation
• Distributed Wasserstein Barycenter
• Data Centric ML
• The Chinese University of Hong Kong

Footnotesedit

1. The topic label follows CUHK IE's official department description, which frames information engineering around information generation, communication, storage, and processing in real-world applications; the ICML 2026 timing in the publications table follows the official ICML 2026 conference page. ↩